I hope you read my last post about what’s wrong with Predictive Analytics – that’s the basis for this post.
I would like to explain how Anticipatory Analytics should work – and give you an idea of what the value is. This is, in essence, what predictive analytics should’ve been in generation 1.0, and how we evolve from that definition of predictive to today’s model. As I mentioned before, I see predictive as being badly implemented more than anything and am hoping this model can improve and replace those faulty implementations.
The outcome for this model: explore the art of the possible. Let’s start with a theoretical example to illustrate better the art of the possible.
A consumer purchases an airline ticket to a tropical destination. In today’s world of traditional marketing the email confirmation would contain links to typical tourist attractions with whom the airline has established relationships. If the consumer buys, they get a percentage of the purchase price – but the number of consumers that take those offers is extremely low.
Any incentive to purchase, or any special offer, will not be customized to that specific consumer to augment the chance of purchase.
Fast forward to predictive analytics events, and the airline may do better than just offer a bunch of random attractions, they might even filter by age and gender of the consumer – or even look at past events and see if they opted for one of the previous offers and then make the offer to the consumer. Since the confidence of these offers is greater, and the recipient is thought to be better known, the offers the consumer gets may be more customized and even addressed to their individual preferences (based on information the airline captured before).
In a world where everything is possible we see a different scenario. The airline would use data from their own repositories, but also from other sources. Compiling information from social channels, communities, partners and alliances, even accessing credit card information from the past they can, close to real time, construct a very effective profile of what offers would or wouldn’t attract the interest of the consumer and extend deeply discounted, but almost guaranteed, offers customized to the preferences of the consumer. Not necessarily based on information that was stored before, but based on analytics conducted ad-hoc on the many data streams related to the consumer.
Each choice the customer would make would then change the potential outcomes of the many other options – which would then be recalculated and the most likely chosen – not the next best, but the next most likely.
In this example we see a consumer that instead of getting an email with 10-12 “opportunities” would get a highly customized package of offers that are almost guaranteed to be interesting and appropriate. Further, if the airline could obtain financial information from a partner about the user, the offers could be of higher or lower value appropriate to each user and further increase the chances of adoption.
Even more interesting, any choice that the consumer makes would alter the calculations for the many other options – in real time resulting in better offers being tendered instantaneously.
This new model, from expecting a consumer to repeat a behavior from someone else in the past to foretelling what a consumer may do based on his or her individual data and needs, and adapting it along the way based on their choices and other data, is the art of the possible today.
Great, you say – so how do I make this work? That’s next week’s third and final post on this series… stay tuned, once more.